With good detail upscaling you don't have to produce as much detail in the base render. you can even use AI based frame completion to lower the calculation complexity. This means that once properly devised and trained you would need less initial 3D data and data throughput and less math to achieve similar results to say a PS5 with a Nintendo 1-3 TFlop device at 360p to 720p mixed resolution(base frame with AI frame completion)->(high grade upscale with detail upscaling)1k->(lower grade upscale)4k at 60Hz is at about the quality level most eyes can't detect much difference past. I think Nvidia need to keep a bit of flexibility with next generation Nintendo to make sure it gets there when it's done no sooner because such a low watt Nintendo could usher a major leap forward in low watt energy efficient gaming. Working with polygon aligned voxelation could help cover enough detail for the upscaler with minimum calc for the base frame. who cares if the base frame looks ugly if the on screen frame looks proper. How to do the illusion of virtual reality very well and very fast is the name of the game along with playability and how easily developers can develop a unique high quality game or 3d video/film.
Here's wishing Nintendo well in their next gen system yet to be as it really could mean a lot.